Assessing the Relevance of Specific Response Features in the Neural Code

Area of Science:

• Neuroscience
• Computational Neuroscience
• Information Theory

Background:

• The neural code describes how nervous systems encode stimuli into neural activity and decode it into responses.
• Information-theoretical methods assess feature relevance by measuring information loss upon feature removal.

Purpose of the Study:

• To compare different information-theoretical methods for assessing neural response feature relevance.
• To identify conditions under which feature relevance assessments converge or diverge.

Main Methods:

• Comparison of various methods for removing response features.
• Analysis of different algorithms for calculating information loss.
• Use of designed examples and analytic derivations.

Main Results:

• Most methods assign different relevance to the same neural response features.
• Differences in relevance are quantitative and qualitative, linked to feature removal and information calculation procedures.
• Conditions for consistent feature relevance ranking or equality were identified, depending on information amount and type.

Conclusions:

• Assessing the relevance of neural response features is more complex than previously assumed.
• Methodological choices significantly impact the identification of relevant features.
• Multiple, method-dependent answers can arise in the quest for relevant neural features.